OpenClaw Adds Guardian Agent For Auto Bash Execution Safety Checks · Digg

/Tech20d ago

OpenClaw Adds Guardian Agent For Auto Bash Execution Safety Checks

3328626192118.8K

VK|@VINCENT_KOC

PS#382|@STEIPETE

Original post

Peter Steinberger 🦞#382

josh avant@joshavant

Now in @openclaw: Use a guardian agent to evaluate the safety of your agent's proposed system calls, only prompting you when it's really needed.

Check out `tools.exec.mode: auto`

Jesse Merhi@jesse_merhi

https://llmgame.scalex.dev go practice for when your security team forces you to approve every bash command.

If only there was a better way… oh wait there is - shipped “auto”’mode for bash executions in @openclaw with @vincent_koc & @joshavant .

https://github.com/openclaw/openclaw/pull/70543

10:12 PM · May 30, 2026 · 34.8K Views

Sentiment

Many users praise OpenClaw's LLM Auto Mode and Guardian Agent for automating security approvals because it provides a safer middle ground than YOLO access, boosts efficiency, and feels more reliable than manual reviews.

Pos

92.8%

Neg

7.2%

8 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Related links

SCALEX.DEVVia

feat(exec): add normalized auto mode by vincentkoc · Pull Request #70543 · openclaw/openclaw

GITHUB.COMVia

safer than yolo auto mode for exec approvals

OPENCLAW.AIVia

Posts from X

Most Activity

Most Activity

VIEWS59KBOOKMARKS91LIKES178RETWEETS15REPLIES21

Vincent Koc@vincent_koc

From YOLO to Auto. LLM-based auto security approvals to make things safer for our users, and works with ANY model. 🦞 https://openclaw.ai/blog/safer-than-yolo-auto-mode-for-exec-approvals

20d|Views 59KLikes 178Bookmarks 91

Kevin S Lin@kevins8

as models continue to get smarter, the future of security approvals is going to involve having models in the loop. yolo is growing up!

Vincent Koc@vincent_koc

From YOLO to Auto. LLM-based auto security approvals to make things safer for our users, and works with ANY model. 🦞 https://openclaw.ai/blog/safer-than-yolo-auto-mode-for-exec-approvals

20d|Views 25KLikes 35Bookmarks 6

Marcin Stachowiak@hi_marcin

The interesting risk: a model approving another model’s actions shares the same failure modes. If the agent can be prompt-injected, so can the approver. Feels less like removing risk and more like moving it up a level. Curious how openclaw handles correlated failure between the two

20d|Views 63

Sedrak Hovhannisyan@Sedrak1010

@joshavant @openclaw Ffc

20d|Views 57Bookmarks 1

Omar Shahine@OmarShahine

@joshavant @openclaw Yes 🙌

20d|Views 125Likes 2

Thomas Tao@Thomas_Tao_1

@vincent_koc The edge cases are where this gets real. Auto approvals feel great until one weird policy boundary shows up and trust drops fast.

20d|Views 4

Vincent Koc@vincent_koc

@ErRahul337 @openclaw Most frontier lab models like GPT5.5 and Opus are trained on thier equivalent of "auto". If you are using a cutting edge model it should be pretty stable.

If unsure it will come back to its owner and ask rather than guessing.

20d|Views 56Likes 1

Kevin S Lin@kevins8

defense in depth. prompt injection isn’t a problem that is going away. a model of equivalent or greater intelligence whose only job is to classify approvals and no extra tools/mcps is going to be much more effective then the model that is already context polluted by the conversation history. if classification is too hard, it can still escalate to a human

20d|Views 55Likes 1

Vincent Koc@vincent_koc

one line to get up and running: openclaw config set tools.exec.mode auto

set a custom model with: openclaw config set tools.exec.reviewer.model openai/gpt-5.5

20d|Views 38

Gonzalo Delgado@gonzalodelgado

@joshavant @openclaw and who guards the guardian agent? 🤔

20d|Views 10Likes 1

Archi cat@Archi_____cat

@vincent_koc Very cool agent, I really like him.

20d|Views 32

Erin Connors@ErinJConnors

@vincent_koc Approval fatigue is how you click yes to the thing you meant to read. Auto is the sane middle: boring commands pass, cwd/argv/env-key weirdness gets a reviewer, and the human only gets woken up when it looks off. YOLO is not a product strategy.

20d|Views 27

BlueprintOS 🧢🦞@blueprint_os

@joshavant @openclaw Safety review on system calls is the right place to add intelligence. We have learned blanket prompts create fatigue, but no review at all is not workable once agents do real ops work.

20d|Views 23

Michael Wall@sound4movement

@vincent_koc @openclaw it’s amazing, as auto-review becomes more robust…I have a hard time letting go of YOLO access…never expected that

20d|Views 22

Alpha Batcher@alphabatcher

@vincent_koc happy to working with OpenClaw everyday

20d|Views 5Likes 1

netrunner@netrunner_btc

@vincent_koc please dont make me trust a sleepy reviewer model

20d|Views 4Likes 1

Relax_to_Rich@R2R_Capital

@vincent_koc Automated risk control boosts efficiency and works with various tech systems.

20d|Views 7

rahul@ErRahul337

@vincent_koc @openclaw Auto mode looks like a smart middle ground between YOLO and constant prompts

How well does the reviewer model handle edge cases or ambiguous commands?

20d|Views 6

Gareth Dwyer ⠕@sixhobbits

@joshavant @openclaw https://en.wikipedia.org/wiki/Quis_custodiet_ipsos_custodes%3F

20d|Views 5

Vincent Koc@vincent_koc

@Thomas_Tao_1 Still better than YOLO, same approach adopted by @AnthropicAI and @OpenAI but we bring this to all models.

20d|Views 4

Load more posts

3328626192118.8K